|
Abstract:
|
When does an effect estimate have a valid causal interpretation? Answering this question requires you to carefully evaluate if the estimation method used is appropriate given the data generating process. While these considerations are familiar when analyzing data from designed experiments, they are often ignored and can be much more challenging when analyzing observational data. This course introduces commonly used methods for estimating dichotomous treatment effects from observational data and tools for evaluating the conditions under which the effect estimate has a valid causal interpretation. In particular, for the estimation of treatment effects this course discusses the use of propensity score matching, inverse probability weighting, and doubly robust methods. For the evaluation of if a causal interpretation is valid for an estimated effect, this course reviews the role of directed graphs as a tool to represent the data generating process, reason about sources of association and bias, and construct a valid estimation strategy. From planning to analysis, these tools provide a rigorous and comprehensive workflow for causal effect estimation from observational data or data with imperfect randomization. This course provides a brief review of the theory behind these estimation and graphical methods and focuses on illustrating their application with a number of examples using some relatively new procedures in SAS/STAT® software. No prior experience with these estimation and graphical methods is assumed.
|
